How It Works
Execution Governance by Design
AICE enforces execution authority, not intelligence. Cognitive systems are free to reason, plan, simulate, optimize, generate code, and propose actions without restriction. Execution occurs only when authority has been explicitly granted.
System Boundary
AICE operates entirely within the system boundary but outside the cognitive domain. There is no cognition-to-execution path. All proposed actions are treated as non-executable until authorized.
Cognitive Proposal
Intelligence produces proposals that carry zero authority. Proposals may include plans, predictions, simulations, or generated code, but cannot directly cause execution.
Structural Authority Evaluation
Each proposed action is evaluated structurally against policy, system state, irreversibility constraints, and execution envelope definitions. Risk influences execution envelope scope only and does not affect authority.
Authorization Artifact
When conditions are satisfied, authorization is issued as a new, independent artifact. Authorization cannot be inferred, learned, optimized, or self-generated by intelligence.
Controlled Execution
Authorized actions execute within a sandboxed environment defined by the execution envelope. All execution crosses an irreversible boundary and is subject to structural constraints. No action is pre-authorized in a way that bypasses authority.

Human Role
What AICE Does Not Do
Result
Relationship to Doctrine
This page describes how the doctrine is enforced. The formal doctrine and invariants are defined in the:
